GenZAgents for Google Gemini — capture conversations from the Gemini web app

Why Gemini is the awkward third provider

Anthropic ships MCP. OpenAI tolerates browser extensions (and has the export-data flow). Google's Gemini is the most closed of the three: no MCP, no public extension API for the consumer Gemini web app. So we capture two ways — a paste-and-submit flow at /sessions/new for ad-hoc capture, and a Workspace-admin export adapter for orgs on Gemini for Workspace. The integration is less elegant than Claude's; it works.

Paste flow for /sessions/new

Engineer finishes a Gemini conversation → opens /sessions/new → pastes the conversation text or the shared-link URL → chooses project + agent → submits. The session is parsed server-side and turned into a receipt with project, agent DID, model (gemini-1.5-pro / gemini-2.0-flash), and runtime cost (best-effort estimate from token counts). The receipt feed treats it identically to an MCP-captured one.

Workspace export adapter for orgs

Gemini for Workspace admins can export per-user Gemini history via the Workspace admin console. Our /api/import/gemini-workspace endpoint accepts the export bundle and creates one receipt per conversation, tagged with the user's Workspace identity (mapped to human_id), the project, and the model. This is how orgs running Gemini at the Workspace tier capture historical AI activity without per-engineer config.

Cross-provider handoff Gemini ↔ Claude ↔ ChatGPT

The same restore_chat / portable-manifest flow works for Gemini. A 4-hour Gemini session becomes a portable JSON that Claude or ChatGPT can ingest as system prompt + memory snapshot. The cross-provider portability is the entire point of having an LLM-neutral receipt format.

What we don't capture today

Live Gemini conversations aren't captured automatically — we don't ship a Gemini browser extension yet (planned for v0.8). Until then, capture is end-of-conversation paste or admin export. If you need live capture across a 100-person Gemini team, talk to us about a custom Workspace add-on (Enterprise tier).

Cost estimation for Gemini receipts

Google's public pricing for gemini-1.5-pro and gemini-2.0-flash is in the SDK price table. The receipt's runtime_cost field is computed from input/output tokens × the listed per-1k rate at receipt-issuance time. The estimate is within 5% of the actual Workspace invoice line item in most cases; the receipt field is marked source=estimate so your finance team knows the difference.

Install

No install. Go to https://genzagents.com/sessions/new → paste the Gemini conversation URL or text → choose project + agent → submit.

What we capture

End-of-conversation capture via paste flow; bulk capture via Workspace admin export.

Verify it works

Paste a Gemini conversation at /sessions/new → submit → confirm receipt on /dashboard.

Common questions

When will the live Gemini browser extension ship?▾

v0.8 (target: August 2026). The blocker is Gemini's DOM stability — Google ships UI changes monthly and our extension would break weekly. We're waiting for the DOM to stabilise or for Google to ship a stable extension API.

Does the Workspace adapter work for Gemini Code Assist?▾

Partially. Code Assist activity shows up in the per-user Workspace export, so we capture the conversation. We don't capture the IDE-specific tool calls (those go through the IDE, not the Gemini chat).

How are Gemini-specific features like grounding citations captured?▾

The grounding-citation block is preserved as a structured field on the receipt (extensions.gemini_grounding). The dashboard shows the citations alongside the response.

What about Gemini Advanced with Deep Research?▾

Captured via paste flow today; the receipt includes the Deep Research source list as extensions.gemini_deep_research_sources. Bulk capture via Workspace export works the same.